Ladder suggestion: Tweak the balancer

[JWhatever]

'ey.

Whenever there are nearly 24 or over players in the server, there tends to form 2 rotations. Each player belongs to one of the rotations and thus plays in every other game.

Suggestion: If there are 24 players (or more) make the balancer divide the players into the top half rated and bottom half rated players. Thus there would be more even games and it wouldn't lead to massive skill differences (team having highest and lowest ranked players of the whole server).

Most of the players who play sub par perks and generally make the game less enjoyable are on the bottom of the rating list. If there were this division, there would be a smaller chance that they'd hamper the top tier games.

A con would be that if a player is under rated, they'd be stuck with the low tier until they get enough points to reach the top tier.


In a nutshell, everyone would still play in every other game, only now their team mates and opponents would be closer to their own rank.


If this actually happened, I'd suggest to put it only on one of the two ball servers. (EU and US #3) while keeping the #4 as a normal.
Speaking of which, bring EU #4 back. You guys sure that lix didn't "accidentally" remove #4 from the server files? It did disappear right after:

Originally Posted by mikesol

All blame is on elixir. Please send him angry pm's demanding a fix.

Division wouldn't work on tbd, not yet at least.


-J

[soccernamlak]

I'm sure Aki or Mikesol can explain the rating system further, but the only thing I could point out is that there are cases where ladder will triple spec someone and others with a triple play in a >24 player situation. So it seems to be taking into account rating as well as spec time/play time.

That being said, I do agree with you that there seems to be generally a 2 squad rotation where player changes are minimal at best. The most obvious solution would be to start a 2nd ladder, but then you run into the potential issue of playing with the same people constantly (but we do that semi-normally anyway...)

Perhaps I can offer a tweak to your system, because I do feel it wouldn't be fair or beneficial to split the lower and higher ratings like that:

In a 24 man rotation, the 6 bottom players are always grouped together for a game and are in Rotation 1. The 6 top players are always grouped together for a game and are in Rotation 2.

The middle 12 are shuffled between Rotation 1 and 2, so that the bottom players will experience relatively high-rated players, and the top players will experience relatively low-rated players.

[Aki1024]

JW: Your idea would take a healthy amount of code to create and test. It isn't out of question, but I'm not sure who would do it. It isn't a bad idea, but it is quite a number of hours to get the idea to work, and further hours to test.

So that more people understand how the parser works, let me try to explain the current one.
Who Joined?
Players a to x join the game. Players y and z are specing. In the previous game, these 12 players played: abehjtvxzAOZ. Since there are 24 players joined into this game (abcdefghijklmnopqrstuvwx) the parser needs to spec 12 players. It gets a list of players in this game that were in last game: abehjtvx (note that z isn't here, as z is specing, and AOZ isn't here since all of them left the server).

Who got Spec'ed
For 8 times, the parser picks a random person from this array and drops them to spec. As 12 specs are needed, all 8 get spec'ed. We now have a game room of cdfgiklmnopqrsuw, and still need to spec 4. For 4 times, 1 player of this set is dropped. Everyone remaining on the list has a 12 in 16, or 75% chance to play. gnqs get spec'ed. The players cdfiklmopruw remain.

What are the teams?
Now we find a balanced team. Ladder gets the ratings of each player, then sorts the players by rating. For ease of example, alphabetical order is rating. The highest two rating players coin flip for which team they are on, then the next two highest rating players coin flip.

First players cd get a coin flip. One player ends up on one team, the other on the other team. Next fi get a coin flip and split to different teams. Same for player pairs kl mo pr and uw. The parser then takes the average (sum/playercount) rating of both teams and compares them via an ELO system, returning a decimal between 0.00 and 1.00 where greater then 0.50 means the left team is favored to win and less then 0.50 means the right team is favored to win. If this number has a spread less then 0.10 (0.45 vs 0.55 is equal to 0.10) then the parser uses these players for the game. If the spread is greater, it remembers this team should this spread be the closest it has yet found.

Assuming the less then 0.10 win chance split isn't found, the above paragraph is repeated up to _20_ times. The parser will try up to 20 times to find a game that has a win spread of less then 0.10. If it can't, it will use the spread that was the closest to 0.10 and runs with it. I don't know if it ever has gotten to 20 times, or how high it has ever gone.

ELO used:
winChanceTeam1 = 1/ (1+10^((team2Rating-team1Rating)/400))
winChanceTeam2 = 1/ (1+10^((team1Rating-team2Rating)/400))

Logic tells me the worse possible team can come from this rank of players:
1, 30, 31, 60, 61, 90, 91, 120, 121, 150, 151, 400
and the coin flips: 1, 31, 61, 91, 121, 151 vs 30, 60, 90, 120, 150, 400
which likely ends up >0.10 spread causing a re-roll until 1 and 400 are on the same team.

Something I just double checked: Player skill order is by rating, not rank. When ladder db is messed up due to double parser and the rating is good but the rank is wrong, the parser is still using the correct number for game creation.

Grammar Nazi Admins are welcome to fix my than then errors